Using model cards for ethical reflection

Abstract

Various representations have been proposed to document machine learning models. In this paper, we analyze how developers make use of one such tool, the Model Card, in ethical reflection. The work is part of a broader research project about epistemic tools for the design of artificial intelligence systems. We conducted a qualitative study based on speculative design sessions. Participants were asked to imagine that they were responsible for the development of an artificial intelligence model in two distinct scenarios: loan applications and university admissions. Regarding Model Cards, the focus of this paper, a thematic analysis of the data suggests that participating developers were selective about which of the ethical issues they reflected upon were actually recorded in their cards. However, participants were hesitant to grant full autonomy to the model they were developing, a contrast with previous studies. These findings may contribute to our current understanding of how developers can leverage epistemic and documentation tools to engage in a more ethically informed design process of artificial intelligence systems.

Ähnliche Arbeiten